Our main contribution to the study of online reputation mechanisms is twofold: First, we analyze panel data in addition to cross-section data. We believe that the dierence between panel and cross-section data is important. In fact, consistently with previous literature, our cross-section results show weak statistical signicance. By contrast, our results from panel data are typically much more signicant, both economically and statistically. We thus agree with Resnick, Zeckhauser, Swanson and Lockwood's [2003] conjecture that there is signicant unobservable seller heterogeneity.